Liver tumour MRI: what do we need for lesion characterization?

Abstract

Background: Hepatic lesions constitute a daily challenge to radiology in clinical settings, and non‐invasive methods are valuable in the characterization of these liver tumours. We undertook our investigation to assess the lesion characterization potential of MRI by evaluating several unenhanced MR sequences and the dynamic gadolinium (Gd)‐enhanced technique. Methods: A total of 116 focal liver lesions in 116 patients were included in our retrospective study, and histological verification was available for 107 lesions. Nine haemangiomas had a follow‐up of 2 years. The 1.5‐T MR system was used. T1‐ and T2‐weighted sequences and dynamic Gd‐enhanced studies were evaluated by two individual readers as separate sequences and also collectively. Lesions were classified into benign or malignant, and a specific diagnosis was proposed. The McNemar test was used in statistical analysis, and interobserver variation was measured using kappa statistics. Results: Lesion classification into benign and malignant tumours (by evaluating all images in concert) was assessed in 83% and 89% of cases by readers 1 and 2, respectively. From single sequences, best lesion classification was achieved with Gd‐enhanced T1 by both readers. The difference in classification was statistically significant when all sequences were evaluated in comparison with any single sequence alone (P = 0.02). Specific diagnosis was correctly determined using all sequences together in 60% and 71% of cases by readers 1 and 2, respectively. For individual sequences, correct diagnosis was most frequently proposed with a Gd‐enhanced T1‐weighted sequence by both readers (59% and 65% for readers 1 and 2, respectively). Conclusion: Multisequential MRI using Gd‐enhanced imaging performs extremely well in liver lesion classification, and with moderate ability to determine a specific diagnosis.